Multidimensional Analysis of Nanoparticles with Highly Disperse

ARTICLE

Multidimensional Analysis of Nanoparticles with Highly Disperse Properties Using Multiwavelength Analytical Ultracentrifugation Johannes Walter,† Konrad Lo¨hr,† Engin Karabudak,‡ Wieland Reis,§ Jules Mikhael,§ Wolfgang Peukert,†,* Wendel Wohlleben,§,* and Helmut Co¨lfen^,* †

Institute of Particle Technology (LFG), Friedrich-Alexander University Erlangen-Nürnberg (FAU), Cauerstr. 4, 91058 Erlangen, Germany, ‡Faculty of Science & Technology, University of Twente, P.O. Box 217, 7500 AE Enschede, The Netherlands, §Material Physics Research, BASF SE, 67056 Ludwigshafen, Germany, and ^ Physical Chemistry, University of Konstanz, Universitätsstr. 10, 78457 Konstanz, Germany

ABSTRACT The worldwide trend in nanoparticle technology toward increasing complexity must be

directly linked to more advanced characterization methods of size, shape and related properties, applicable to many different particle systems in science and technology. Available techniques for nanoparticle characterization are predominantly focused on size characterization. However, simultaneous size and shape characterization is still an unresolved major challenge. We demonstrate that analytical ultracentrifugation with a multiwavelength detector is a powerful technique to address multidimensional nanoparticle analysis. Using a high performance optical setup and data acquisition software, information on size, shape anisotropy and optical properties were accessible in one single experiment with unmatched accuracy and resolution. A dynamic rotor speed gradient allowed us to investigate broad distributions on a short time scale and differentiate between gold nanorod species including the precise evaluation of aggregate formation. We report how to distinguish between different species of single-wall carbon nanotubes in just one experiment using the wavelength-dependent sedimentation coefficient distribution without the necessity of time-consuming purification methods. Furthermore, CdTe nanoparticles of different size and optical properties were investigated in a single experiment providing important information on structure
property relations. Thus, multidimensional information on size, density, shape and optical properties of nanoparticulate systems becomes accessible by means of analytical ultracentrifugation equipped with multiwavelength detection. KEYWORDS: analytical ultracentrifugation (AUC) with multiwavelength detection . speed ramps . particle size distribution . nanoparticles . gold nanorods . carbon nanotubes

T

he accurate evaluation of particle size and shape distributions and optical information in particle technology is essential, both for scientific structure
property investigations and for commercial quality control. Their correct determination is indispensable more than ever because product properties are directly influenced by size and shape, and modern nanoparticle engineering would be impossible without the precise knowledge of the particle characteristics. Nanoparticulate systems have been shown to possess enhanced properties in a manifold manner. For very small semiconductor nanoparticles, quantum dots (QD), it was shown that the electro-optical properties1 are a function of the particle size, making them a promising material in electronic devices and WALTER ET AL.

solar cells.2 Although powerful methods were developed to get access to those systems within the last years,3,4 their true multidimensional characterization is still not possible, and assumptions have to be made regarding the particle size-dependent optical properties. Shape anisotropic structures such as semiconductor or metallic rods and carbon nanotubes have various interesting properties, which can be used for thin film materials and optical applications.5,6 However, the resulting product qualities are strongly dependent upon particle distribution and stability. Further examples include organic nanoparticles, like β-carotene, which have size- and structure-dependent optical properties.7 Moreover, recent research has shown that the toxicology is strongly dependent on these disperse properties,8
11 VOL. 8

’

NO. 9

’

* Address correspondence to [email protected], [email protected], [email protected]. Received for review December 19, 2013 and accepted August 17, 2014. Published online August 17, 2014 10.1021/nn503205k C 2014 American Chemical Society

8871–8886

’

2014

8871 www.acsnano.org

WALTER ET AL.

technique was never applied to complex mixtures to directly correlate hydrodynamic and spectral properties. Although these developments significantly widened the possibilities of nanoparticle analysis, AUC is still a highly underrepresented and rarely applied technique for nanoparticle characterization. One of the major reasons could be the special challenges arising from these systems. Most of the particles in solution have a hybrid structure, in which a core particle is stabilized by attached ions or surfactants against agglomeration. As a result, the total particle density is often unknown and will change as a function of core size. This problem could be recently overcome by a simultaneous evaluation of sedimentation and diffusion coefficients, which yielded the density of monodisperse particles in a multicomponent mixture.27 However, polydisperse particles still remain a challenge. The nature and binding of the surfactants is crucial for colloidal stability and hence many of the methods developed for biological systems cannot be directly transferred to particulate systems. This mainly affects the evaluation of heavy components and broad distributions, as it will be later discussed in our manuscript. Therefore, new methods are required addressing these and other particular challenges. Herein, we describe a MWL-AUC with an improved optical setup and a new data acquisition software. We used preparative ultracentrifuges from Beckman Coulter (model Optima L-90K or Optima XL-80K) as a platform for our developments. These centrifuges allow a minimum rotor speed of 1 krpm and a maximum speed of 60 krpm in combination with the analytical 4-hole rotor. The spectrometer was positioned directly within the optical path, which leads to a substantially increased light sensitivity.23 An UV
vis-enhanced optical fiber grants enhanced long-term stability due to reduced fiber bleaching. An UVenhanced aluminum coated mirror replaced the prism, and the collimation was reworked in such way that a sheet shaped light profile is formed directly after the fiber. A new xenon flash lamp (Hamamatsu L-9455-1x) in combination with a high performance UV
vis CCD array spectrometer (Ocean Optics USB2000þ ES) allows to record full spectra every 3 milliseconds in a high-speed acquisition mode directly on the spectrometer memory. Recent mechanical drawings are provided via the openAUC framework. This setup guarantees a much more uniform intensity profile compared to the previous setups, thus providing a much better signal-to-noise ratio in the UV and NIR, as we will show later. We show that the improved data quality enabled us to perform high-end hydrodynamic analysis. Furthermore, continuous speed ramp experiments could be carried out, which are very useful for polydisperse particles using multiwavelength detection.28 This new multiwavelength speed ramp functionality with a dynamic rotor speed gradient offers the VOL. 8

’

NO. 9

’

8871–8886

’

ARTICLE

and the European regulatory definition of nanomaterials calls for precise size distributions.12 Many characterization techniques such as static and dynamic light scattering, field flow fractionation,13 transmission electron microscopy, and chromatography among other techniques were developed for nanoparticle analysis. However, most of these methods suffer either from the complex inversion of ensemble measurements, the limited applicability for broad distributions or the underlying statistics. The latter is predominantly true for scanning or transmission electron microscopy and atomic force microscopy. Even though these methods provide highly accurate information on size, they are tedious and drying effects can impair the accuracy of particle counting, which is a clear drawback for the analysis of polydisperse samples. Usually, combinations of several time-consuming and expensive methods have to be employed for the evaluation of size, shape parameters, composition and optical characterization. For almost 100 years, analytical ultracentrifugation (AUC) has been shown to be a very powerful and versatile tool for a number of important applications.14 Because of very high rotor speeds up to 60 krpm, even particles with sizes in the lower nanoscale can be measured precisely and with Ångström resolution15 with a variety of detectors such as single wavelength absorption, interference and fluorescence.16 Even though biological systems were predominantly examined in the past, these detectors give access to almost every sample system in solution, e.g., small metal or semimetal clusters,17 polymers and biological macromolecules.20,21 Even very small prenucleation clusters can be successfully separated from sedimenting salt ions.18 Besides size analyses and information on shape anisotropy, samples can be examined with respect to their stability and interactions.9,10 AUC was compared to other techniques in the above list and showed superior resolution for mixtures as well as the correct determination of particle concentrations.19
21 Especially a study on latex mixtures performed worldwide in the laboratories of the Bayer group with various analytical techniques demonstrated the quality of AUC for particle size analysis in comparison to other techniques very well.21 The capabilities of AUC were further enhanced by the development of a multiwavelength UV
vis detector (MWL-AUC), which is capable of determining full UV
vis spectra for each detected species.22,23 This detector was developed on an open source basis24 and it was demonstrated that this detector can resolve complex mixtures and yield information about the sample components.25,26 However, the first generations still had limited data quality, which at its best reached the quality of the commercial AUC, the Beckman-Coulter Optima XL-A/I.23 For that reason as well as due to the limited reproducibility, this

8872

2014 www.acsnano.org

ARTICLE Figure 1. Measuring schemes of two different AUC experiments. (a) Sedimentation velocity experiment with scanning in radial dimension. (b) Speed ramp experiment at fixed radial position with variable rotor speed gradient.

possibility of measuring wide distributions on a very short time scale with high resolution. Using our high performance data acquisition software, the informational content of the measurements was improved significantly due to an enhanced sample capacity and detection capability. We validated all our developments with several polystyrene as well as protein standards and outline its potential for multidimensional particle analysis by means of gold nanorods, CdTe nanoparticles and single-/multi-wall carbon nanotubes. Plans of the setup can be obtained for free from the authors in the framework of the openAUC project (see also http://wiki.bcf2. uthscsa.edu/openAUC/). General schematic diagrams of the modified absorbance optics are also provided in the Supporting Information (SI).24 RESULTS AND DISCUSSION Theoretical Foundations. In principle AUC is based on the optical detection of particle motion during their fractionation caused by a centrifugal field. For a sedimentation velocity experiment, this can be achieved with a detector measuring the flash light intensity as a function of time and space (Figure 1a). The apparent sedimentation coefficient is calculated as follows:
 1 r s ¼ 2 ln (1) ω t rm ω is the angular velocity, t the time, r the radial measuring and rm the radial meniscus position. The sedimentation coefficient has the dimension of time and is usually expressed in Svedberg (1 S equals 10
13 seconds). Besides, we can use Stokes' law to transform the sedimentation coefficient to the equivalent diameter of a sphere with equal sedimentation and density: sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 18ηs d Stokes ¼ (2) Fp
Fs dStokes is the diameter, η the solvent viscosity and F the solvent (s) and particle (p) density. The aforementioned equation only assumes sedimentation neglecting diffusion effects. The Lamm eq 3 gives a more WALTER ET AL.

accurate thermodynamic approach of the centrifugation process.29,30

∂c/∂t is the change in mass concentration with time and D is the translational diffusion coefficient. Depending on the measurement conditions (e.g., rotor speed, particle size, etc.), either the sedimentation or the diffusion is the dominant term. In particular, for particle sizes below 20 nm diffusion leads to a significant broadening of the particle size distribution (PSD), which has to be corrected. The translational diffusion coefficient of a small particle is defined by the Stokes
Einstein equation: D ¼

RT RT ¼ NA f NA 3πηd H

(4)

R is the gas constant, T the temperature in Kelvin, NA the Avogadro's number, f the translational frictional coefficient and dH the hydrodynamic diameter. Direct boundary modeling methods apply the mass conservation approach to the experimental data and evaluate sedimentation as well as diffusion. Eq 5 illustrates the hydrodynamic dependency of these values for a particle of arbitrary shape: D(s, f =f 0 ) ¼ RT[2(Fp
Fs )]1=2 [18s1=2 NA π(f =f 0 η)3=2 ]
1 (5) f/f0 is the frictional ratio, which is the friction factor divided by the friction factor of a sphere of equal volume. This ratio can be used to estimate shape parameters of particles in the case that the solvent and particle parameters are known. For various systems models were developed to characterize the dimensions and aspect ratios of rods, discs and plates based upon their friction.31
34 Using the Kirkwood approximation and bead modeling it becomes even possible to calculate the friction coefficients for arbitrary shapes.35,36 However, it has to be kept in mind that the frictional ratio represents in principle an infinite number VOL. 8

’

NO. 9

’

8871–8886

’

8873

2014 www.acsnano.org

WALTER ET AL.

unknown density/viscosity profiles. The latter is also problematic for systems that may specifically interact with the gradient liquids. The commercial analytical ultracentrifuges from Beckman Coulter achieve rotor speeds up to 60 krpm but do not give the possibility of speed ramps. A turbidity detector designed by the user community transferred this technique to the AUC and even though the particle size range could be extended by higher rotor speeds, the spectral dimension stayed inaccessible.28,38,39 Herein, we implemented this indispensable function to our MWL-AUC giving numerous additional possibilities of sample analysis. It has to be noted that a 2-dimensional diffusion/shape analysis is not possible for a speed ramp experiment because the radial dimension is not recorded in such experiments. For absorbance measurements, the measured attenuation can be expressed according to Lambert
Beer's law (within the linear range of concentration):
 I0 ln ¼ εcl (6) I

ARTICLE

of different shapes that can have the same friction. Nevertheless, for systems of known geometry such as carbon nanotubes or nanorods, the frictional ratio can be used to estimate shape parameters. The measurement and evaluation of very broad distributions poses considerable challenges for a classical AUC experiment since the sedimentation velocity scales with the square of the particle size. A multispeed analysis developed by W. Stafford is able to address this need because one sample is measured incrementally at multiple rotor speeds.37 The sedimentation data from each speed are then combined into a single continuous distribution function. However, this is a very time-consuming procedure. Moreover, our experiments indicated that the stability of the sedimentation boundary is highly sensitive to the stepwise ramping for rotor speeds below 10 krpm. Furthermore, the evaluation of fast sedimenting species such as heavy particles is hardly possible in a sedimentation velocity experiment due to the limited temporal resolution (the particle sediments faster than the step motor is able to scan). For such systems it is the current stateof-the-art to change the solvent viscosity and/or density to slow down the sedimentation. This can either be achieved by ionic and nonionic additives or by using a more dense solvent such as ethylene glycol or glycerol. Although this is applicable to various macromolecules, for many particulate systems this is not an option. The stability of particles in solution is strongly dependent upon the interparticle forces. Specific ligands stabilize particles against agglomeration or particles are intrinsically stable due their surface chemistry. Thus, even a slight change of the solvent such as its polarity can lead to instable dispersions. For systems where primary particles as well as the degree of agglomeration are studied this is an inacceptable drawback. An exception are deuterated solvents because they normally do not affect the colloidal stability. However, for very dense particles the corresponding change in sedimentation velocity is negligible. By keeping the measurement position constant no sedimenting species is missed and the examined distribution can be as broad as desired by varying the rotor speed gradient and the measurement duration (Figure 1b). In addition, lower rotor speeds extend the measuring range to particles of unit density (e.g., polystyrene) up to 3 μm without changing the solvent conditions.28 Disc centrifuges use this technique to measure PSDs, but they are limited to a single wavelength and their maximum speed is 24 krpm (10% maximum gravitational force of an AUC). Therefore, the smallest particle size is restricted to a few 10 nm. Moreover, disc centrifugation requires a standard sample of known size and density to be used for calibration. It cannot provide fundamental properties such as the sedimentation coefficient due to a lack of thermal control and the use of gradient liquids of

I0/I is the ratio of the light intensities of solvent and solution, ε is the extinction coefficient, c is the mass concentration and l is the path length. The extinction coefficient can be calculated based on Mie's scattering theory, which is applicable in a strict sense for spherical particles only.40 It is dependent upon the extinction efficiency, the particle size as well as its density.41 For absorbing particles, the absorbance as well as the scattering contributions have to be taken into account.42 To obtain the correct concentrations for each species and distribution, the particle size and wavelength-dependent scattering has to be considered. The influence of Mie correction on size distribution was assessed in detail by M. D. Lechner.42,43 For the time being, we only added scattering to the evaluation of the speed ramp data, since the only evaluation software available for speed ramp data by M. D. Lechner includes a scattering correction based on Mie theory by default. Signal-to-Noise Performance. As an extension to the previous setup two or more flashes were averaged to keep the noise introduced by the flash-lamp and spectrometer as low as possible. A dark current correction was integrated to grant drift free operation. All absorbance data was calculated as follows: Either the radial intensity profile of a reference channel or the air above the sample's meniscus was used after the experiment to calculate an average intensity value for each run. This value was then used to calculate the absorbance. The last case led to pseudoabsorbance distributions. This procedure improved our signalto-noise ratio (SNR) significantly because any intensity fluctuations of the reference were not taken into account. A wavelength averaging functionality was implemented to reduce the readout noise of the spectrometer and to provide an even better SNR. VOL. 8

’

NO. 9

’

8871–8886

’

8874

2014 www.acsnano.org

WALTER ET AL.

ARTICLE

In the data acquisition software a desired number of pixels can be specified for averaging. Herein, we averaged three pixels resulting in a theoretical spectral resolution of 1 nm. Since the absolute resolution defined by the diffraction grading is approximately 2 nm, up to 6 pixels can be averaged without losing any resolution. In Figure 2 the largely reduced intrinsic detector noise of our improved MWL-AUC is shown by comparing the results of a radial scan at 650 nm (an empty rotor hole) with the commercially available single wavelength absorption detection of the XL-A from Beckman Coulter. Table 1 demonstrates the uniform detector noise of the MWL-AUC design in whole accessible spectrum, whereas the XL-A has a low SNR in the UV range only. The noise contribution is strongly dependent on the amount of light getting to the detector. The XL-A with its free beam arrangement is optimized to get a maximum signal in the UV range, which is mainly important for biological applications. For particulate systems, where the visible range is quite often important, this is a serious disadvantage due to the increased noise contribution there. For the MWL-design we observe a slightly higher standard deviation in the UV range due to the fiber coupling, but it performs better by a factor of 3 in the visible range compared to the XL-A. Therefore, our design is superior to the commercial AUC calling for manifold experimental studies including measurements in the UV, visible as well as NIR. Later on, this improvement will be highly beneficial for our studies on the multi-wall carbon nanotubes (mwCNTs). Validation. The data acquisition (DAQ) software and the accuracy of the MWL-AUC were validated by means of Duke NIST traceable polystyrene (PS) standards and two standard proteins in sedimentation velocity experiments. To assess the quality of our MWL-AUC design several parameters can be applied such as the mass averaged diameter or the width of the distribution (e.g., d90/d10 or d75/d25). The weighting of the distribution (number, size, area, mass or intensity) is dependent upon the measurement technique. A different weighting will therefore lead to a different mean diameter. For absorbance based AUC data, the given values are mass weighted for small particles (